Rollup merge of rust-lang#64840 - michaelwoerister:self-profiling-rai…

…i-refactor, r=wesleywiser

SelfProfiler API refactoring and part one of event review

This PR refactors the `SelfProfiler` a little bit so that most profiling methods are RAII-based. The codegen backend code already had something similar, this refactoring pulls this functionality up into `SelfProfiler` itself, for general use.

The second commit of this PR is a review and update of the existing events we are already recording. Names have been made more consistent. CGU names have been removed from event names. They will be added back in when function parameter recording is implemented.

There is still some work to be done for adding new events, especially around trait resolution and the incremental system.

r? @wesleywiser

src/librustc/session/mod.rs

@@ -32,7 +32,7 @@ use syntax::source_map;
 use syntax::parse::{self, ParseSess};
 use syntax::symbol::Symbol;
 use syntax_pos::{MultiSpan, Span};
-use crate::util::profiling::SelfProfiler;
+use crate::util::profiling::{SelfProfiler, SelfProfilerRef};
 
 use rustc_target::spec::{PanicStrategy, RelroLevel, Target, TargetTriple};
 use rustc_data_structures::flock;
@@ -129,7 +129,7 @@ pub struct Session {
     pub profile_channel: Lock<Option<mpsc::Sender<ProfileQueriesMsg>>>,
 
     /// Used by `-Z self-profile`.
-    pub self_profiling: Option<Arc<SelfProfiler>>,
+    pub prof: SelfProfilerRef,
 
     /// Some measurements that are being gathered during compilation.
     pub perf_stats: PerfStats,
@@ -835,24 +835,6 @@ impl Session {
         }
     }
 
-    #[inline(never)]
-    #[cold]
-    fn profiler_active<F: FnOnce(&SelfProfiler) -> ()>(&self, f: F) {
-        match &self.self_profiling {
-            None => bug!("profiler_active() called but there was no profiler active"),
-            Some(profiler) => {
-                f(&profiler);
-            }
-        }
-    }
-
-    #[inline(always)]
-    pub fn profiler<F: FnOnce(&SelfProfiler) -> ()>(&self, f: F) {
-        if unlikely!(self.self_profiling.is_some()) {
-            self.profiler_active(f)
-        }
-    }
-
     pub fn print_perf_stats(&self) {
         println!(
             "Total time spent computing symbol hashes:      {}",
@@ -1251,7 +1233,7 @@ fn build_session_(
         imported_macro_spans: OneThread::new(RefCell::new(FxHashMap::default())),
         incr_comp_session: OneThread::new(RefCell::new(IncrCompSession::NotInitialized)),
         cgu_reuse_tracker,
-        self_profiling: self_profiler,
+        prof: SelfProfilerRef::new(self_profiler),
         profile_channel: Lock::new(None),
         perf_stats: PerfStats {
             symbol_hash_time: Lock::new(Duration::from_secs(0)),

src/librustc/ty/context.rs

@@ -45,6 +45,7 @@ use crate::ty::CanonicalPolyFnSig;
 use crate::util::common::ErrorReported;
 use crate::util::nodemap::{DefIdMap, DefIdSet, ItemLocalMap, ItemLocalSet};
 use crate::util::nodemap::{FxHashMap, FxHashSet};
+use crate::util::profiling::SelfProfilerRef;
 
 use errors::DiagnosticBuilder;
 use arena::SyncDroplessArena;
@@ -1030,6 +1031,8 @@ pub struct GlobalCtxt<'tcx> {
 
     pub dep_graph: DepGraph,
 
+    pub prof: SelfProfilerRef,
+
     /// Common objects.
     pub common: Common<'tcx>,
 
@@ -1260,6 +1263,7 @@ impl<'tcx> TyCtxt<'tcx> {
             arena: WorkerLocal::new(|_| Arena::default()),
             interners,
             dep_graph,
+            prof: s.prof.clone(),
             common,
             types: common_types,
             lifetimes: common_lifetimes,

src/librustc/ty/query/plumbing.rs

@@ -112,7 +112,7 @@ impl<'a, 'tcx, Q: QueryDescription<'tcx>> JobOwner<'a, 'tcx, Q> {
             let mut lock = cache.get_shard_by_value(key).lock();
             if let Some(value) = lock.results.get(key) {
                 profq_msg!(tcx, ProfileQueriesMsg::CacheHit);
-                tcx.sess.profiler(|p| p.record_query_hit(Q::NAME));
+                tcx.prof.query_cache_hit(Q::NAME);
                 let result = (value.value.clone(), value.index);
                 #[cfg(debug_assertions)]
                 {
@@ -128,7 +128,7 @@ impl<'a, 'tcx, Q: QueryDescription<'tcx>> JobOwner<'a, 'tcx, Q> {
                             // in another thread has completed. Record how long we wait in the
                             // self-profiler.
                             #[cfg(parallel_compiler)]
-                            tcx.sess.profiler(|p| p.query_blocked_start(Q::NAME));
+                            tcx.prof.query_blocked_start(Q::NAME);
 
                             job.clone()
                         },
@@ -170,7 +170,7 @@ impl<'a, 'tcx, Q: QueryDescription<'tcx>> JobOwner<'a, 'tcx, Q> {
             #[cfg(parallel_compiler)]
             {
                 let result = job.r#await(tcx, span);
-                tcx.sess.profiler(|p| p.query_blocked_end(Q::NAME));
+                tcx.prof.query_blocked_end(Q::NAME);
 
                 if let Err(cycle) = result {
                     return TryGetJob::Cycle(Q::handle_cycle_error(tcx, cycle));
@@ -382,8 +382,9 @@ impl<'tcx> TyCtxt<'tcx> {
         }
 
         if Q::ANON {
+
             profq_msg!(self, ProfileQueriesMsg::ProviderBegin);
-            self.sess.profiler(|p| p.start_query(Q::NAME));
+            let prof_timer = self.prof.query_provider(Q::NAME);
 
             let ((result, dep_node_index), diagnostics) = with_diagnostics(|diagnostics| {
                 self.start_query(job.job.clone(), diagnostics, |tcx| {
@@ -393,7 +394,7 @@ impl<'tcx> TyCtxt<'tcx> {
                 })
             });
 
-            self.sess.profiler(|p| p.end_query(Q::NAME));
+            drop(prof_timer);
             profq_msg!(self, ProfileQueriesMsg::ProviderEnd);
 
             self.dep_graph.read_index(dep_node_index);
@@ -451,9 +452,8 @@ impl<'tcx> TyCtxt<'tcx> {
         // First we try to load the result from the on-disk cache.
         let result = if Q::cache_on_disk(self, key.clone(), None) &&
                         self.sess.opts.debugging_opts.incremental_queries {
-            self.sess.profiler(|p| p.incremental_load_result_start(Q::NAME));
+            let _prof_timer = self.prof.incr_cache_loading(Q::NAME);
             let result = Q::try_load_from_disk(self, prev_dep_node_index);
-            self.sess.profiler(|p| p.incremental_load_result_end(Q::NAME));
 
             // We always expect to find a cached result for things that
             // can be forced from `DepNode`.
@@ -469,21 +469,17 @@ impl<'tcx> TyCtxt<'tcx> {
 
         let result = if let Some(result) = result {
             profq_msg!(self, ProfileQueriesMsg::CacheHit);
-            self.sess.profiler(|p| p.record_query_hit(Q::NAME));
-
             result
         } else {
             // We could not load a result from the on-disk cache, so
             // recompute.
-
-            self.sess.profiler(|p| p.start_query(Q::NAME));
+            let _prof_timer = self.prof.query_provider(Q::NAME);
 
             // The dep-graph for this computation is already in-place.
             let result = self.dep_graph.with_ignore(|| {
                 Q::compute(self, key)
             });
 
-            self.sess.profiler(|p| p.end_query(Q::NAME));
             result
         };
 
@@ -551,7 +547,7 @@ impl<'tcx> TyCtxt<'tcx> {
                 key, dep_node);
 
         profq_msg!(self, ProfileQueriesMsg::ProviderBegin);
-        self.sess.profiler(|p| p.start_query(Q::NAME));
+        let prof_timer = self.prof.query_provider(Q::NAME);
 
         let ((result, dep_node_index), diagnostics) = with_diagnostics(|diagnostics| {
             self.start_query(job.job.clone(), diagnostics, |tcx| {
@@ -571,7 +567,7 @@ impl<'tcx> TyCtxt<'tcx> {
             })
         });
 
-        self.sess.profiler(|p| p.end_query(Q::NAME));
+        drop(prof_timer);
         profq_msg!(self, ProfileQueriesMsg::ProviderEnd);
 
         if unlikely!(self.sess.opts.debugging_opts.query_dep_graph) {
@@ -619,7 +615,7 @@ impl<'tcx> TyCtxt<'tcx> {
             let _ = self.get_query::<Q>(DUMMY_SP, key);
         } else {
             profq_msg!(self, ProfileQueriesMsg::CacheHit);
-            self.sess.profiler(|p| p.record_query_hit(Q::NAME));
+            self.prof.query_cache_hit(Q::NAME);
         }
     }